This is my full course for the AWS Certified Solutions Architect - Associate (SAA-C03) exam: the video above (1 hour 34 minutes) and the written notes below, domain by domain. I checked both against the official exam guide, which AWS last revised on 29 October 2026 (version 1.1).
Watch the course, then use the notes to revise. The chapters jump to each domain:
- 1:16 Domain 1: Design Secure Architectures
- 28:07 Domain 2: Design Resilient Architectures
- 51:56 Domain 3: Design High-Performing Architectures
- 1:13:28 Domain 4: Design Cost-Optimized Architectures
- 1:29:32 Four mistakes to avoid and exam day tactics
SAA-C03 exam at a glance
| Item | Detail |
|---|---|
| Exam code | SAA-C03 |
| Questions | 65 (50 scored, 15 unscored) |
| Time | 130 minutes |
| Passing score | 720 out of 1,000 (scaled) |
| Cost | 150 USD |
| Question types | Multiple choice (one correct answer) and multiple response (two or more correct answers out of five or more options) |
| Where you take it | Pearson VUE testing centre or online proctored |
| Who it is for | People in a solutions architect role, with at least 1 year of hands-on experience designing cloud solutions on AWS |
| Exam guide version | 1.1, published 29 October 2026 |
Three scoring rules matter for how you sit the exam:
- There is no penalty for guessing. An unanswered question is scored as incorrect, so never leave one blank.
- The 15 unscored questions are not marked. AWS uses them to test future questions, and you cannot tell which ones they are.
- You pass on the exam as a whole. AWS uses compensatory scoring, so you do not need a passing score in each domain.
The 65 questions in 130 minutes work out at two minutes a question.
See the study system in action
How StudyTech AI works
Watch how your answers become a focused daily plan, targeted practice and flashcards. See how StudyTech AI tracks your progress before you decide when to book.
Ready to explore? Visit StudyTech AI to see supported certifications and choose your next step.
Explore StudyTech AIWhat changed for SAA-C03 in 2026
The exam code and the four domain weights are the same. The service list and several AWS service names are not.
| Change | Date | What it means for you |
|---|---|---|
| Exam guide version 1.0 published | 28 August 2026 | Study from the current guide, not an older PDF |
| Exam guide version 1.1 published | 29 October 2026 | Amazon Elastic Transcoder was removed from the in-scope services list |
| Amazon QuickSight became part of Amazon Quick | October 2025 | The exam guide lists Amazon Quick. AWS renamed it AWS Quick on 9 November 2026, and QuickSight is the Quick Sight capability inside it |
| AWS Snowball Edge closed to new customers | 7 November 2025 | AWS Snow Family is still in scope, but AWS points new customers to AWS DataSync for online transfers |
| SSE-C switched off by default on new S3 buckets | April 2026 | Server-side encryption with customer-provided keys now has to be turned on deliberately |
| Regional NAT gateways launched | November 2025 | One NAT gateway can now span Availability Zones. The exam guide still frames the choice as one shared NAT gateway or one per Availability Zone |
Many courses and practice exams were recorded before these changes. If your material still teaches Elastic Transcoder or calls the dashboard service QuickSight throughout, it has not been updated.
The four SAA-C03 domains and their weights
| Domain | Weight | Official task statements |
|---|---|---|
| 1. Design Secure Architectures | 30% | 1.1 Design secure access to AWS resources. 1.2 Design secure workloads and applications. 1.3 Determine appropriate data security controls |
| 2. Design Resilient Architectures | 26% | 2.1 Design scalable and loosely coupled architectures. 2.2 Design highly available and/or fault-tolerant architectures |
| 3. Design High-Performing Architectures | 24% | 3.1 Storage. 3.2 Compute. 3.3 Database. 3.4 Network architectures. 3.5 Data ingestion and transformation |
| 4. Design Cost-Optimized Architectures | 20% | 4.1 Storage. 4.2 Compute. 4.3 Database. 4.4 Network architectures |
Security and resilience are 56% of the scored content. IAM, VPCs, load balancing, decoupling and disaster recovery deserve more of your time than a service you might see once.
Domain 1: Design Secure Architectures (30%)
Domain 1 is the largest. It covers who can access your resources, how you manage many accounts, how you lock down a network, the services that stop attacks, and how you encrypt and protect data. Many questions give a security requirement and ask for the option with the least access or the least effort.
Two foundations: global infrastructure and shared responsibility
- Regions and Availability Zones. A Region is a geographic area made up of several Availability Zones (AZs). An AZ is one or more separate data centres. High availability answers spread resources across multiple AZs. Disaster recovery answers use multiple Regions.
- The shared responsibility model. AWS is responsible for security of the cloud: data centres, hardware and the software that runs its services. You are responsible for security in the cloud: your data, access, network settings and encryption. The more managed the service, the less you patch. You patch the operating system on Amazon EC2. AWS does it on AWS Lambda and Amazon RDS.
IAM: users, groups, roles and policies
| Building block | What it is |
|---|---|
| IAM user | An identity for one person or application, with long-term credentials |
| IAM group | A collection of users that share permissions |
| IAM role | An identity with no long-term credentials. A person or service assumes it and gets temporary credentials |
| IAM policy | A JSON document listing what is allowed or denied, attached to users, groups or roles |
The rules the exam tests repeatedly:
- Least privilege. Grant only the permissions the job needs. A broad AWS managed policy such as read-only access to all of Amazon S3 is a trap when the application needs one bucket.
- Roles for applications. When an application on EC2 needs to read from S3, attach an IAM role to the instance. Never store access keys on the server or in code.
- Protect the root user. Turn on multi-factor authentication, create no access keys for it, and use it only for the few tasks that need it.
- Identity-based or resource-based. An identity-based policy says what a user, group or role can do. A resource-based policy, such as an S3 bucket policy, says who can access that resource, including another AWS account.
- An explicit deny always wins. If any policy denies an action, no number of allows overrides it.
Roles, STS and cross-account access
Roles get temporary credentials from the AWS Security Token Service (STS). For cross-account access, account A creates a role that trusts account B, and a user in account B assumes it. If a question asks how one account can access another without sharing long-term credentials, the answer is a cross-account IAM role.
Managing multiple accounts
- AWS Organizations groups accounts into organisational units (OUs) with one consolidated bill.
- Service control policies (SCPs) set the maximum permissions for an OU or account, including the root user of each member account. An SCP grants nothing by itself. It only limits.
- AWS Control Tower sets up a governed multi-account environment called a landing zone, with ready-made rules called controls (formerly guardrails).
- AWS Resource Access Manager (AWS RAM) shares resources such as subnets or transit gateways between accounts.
Federation and identity services
| Who is signing in | Service |
|---|---|
| Employees signing in to AWS accounts | AWS IAM Identity Center (single sign-on across the organisation) |
| An existing Microsoft Active Directory | AWS Directory Service: AWS Managed Microsoft AD, or AD Connector as a proxy to the on-premises directory |
| Customers of your web or mobile app | Amazon Cognito: a user pool for sign-up and sign-in, an identity pool for temporary AWS credentials |
With federation, nobody gets an IAM user. People sign in with a company identity and are given a role for the session. An answer that creates an IAM user for every employee is almost always wrong.
Securing the network with a VPC
Amazon VPC is your private network in AWS, divided into subnets that each sit in one AZ.
- Public subnet: has a route to an internet gateway. Load balancers go here.
- Private subnet: has no route to the internet gateway. Application servers and databases go here.
- Route table: decides where traffic leaving a subnet goes. It is what makes a subnet public or private.
- NAT gateway: lets private servers reach out to the internet while nothing can start a connection back in.
The expected pattern is a multi-tier architecture: a load balancer in public subnets, application servers in private subnets and the database in its own private subnets, across at least two AZs.
| Security group | Network ACL | |
|---|---|---|
| Protects | One resource | A whole subnet |
| Rules | Allow only | Allow and deny, checked in number order |
| State | Stateful: responses are allowed automatically | Stateless: return traffic must be allowed separately |
| Can block one IP address | No | Yes |
A security group rule can reference another security group, so a database can accept traffic only from the application servers' group. To reach a private instance with no inbound ports, SSH keys or bastion host, use AWS Systems Manager Session Manager.
Private connections
| Need | Answer |
|---|---|
| Private servers reaching S3 or Amazon DynamoDB without a NAT gateway | Gateway endpoint (free, added as a route) |
| Private access to most other AWS services | Interface endpoint, powered by AWS PrivateLink (charged per hour and per gigabyte) |
| Quick, low-cost encrypted link to a data centre | AWS Site-to-Site VPN |
| Dedicated line with consistent performance | AWS Direct Connect (takes weeks, not encrypted by default) |
| Individual employees connecting from laptops | AWS Client VPN |
Protecting applications from attacks
| Service | Job |
|---|---|
| AWS WAF | Web application firewall: blocks SQL injection, cross-site scripting and bad bots, and rate limits requests |
| AWS Shield | DDoS protection. Standard is free and automatic. Advanced is paid, with a response team and cost protection |
| AWS Firewall Manager | Rolls out WAF rules, Shield Advanced and security groups across every account |
| AWS Network Firewall | Filters traffic at the VPC level |
| Amazon GuardDuty | Threat detection: something suspicious is happening |
| Amazon Inspector | Vulnerability scanning: something could be exploited |
| Amazon Macie | Finds sensitive data in S3 buckets |
| Amazon Detective | Investigates a finding after the fact |
| AWS Security Hub | Pulls findings from all of these into one place |
Credentials, encryption and key management
- AWS Secrets Manager stores secrets and rotates them automatically, including RDS database passwords. If the question says automatic rotation, this is the answer.
- AWS Systems Manager Parameter Store stores configuration and secrets. Its standard tier is free, but it does not rotate secrets for you.
- AWS KMS creates and manages encryption keys. An AWS managed key is run by AWS for a service. A customer managed key is yours: you set the key policy, turn on rotation and can disable it. If the company must control who uses the key, choose a customer managed key.
- Envelope encryption. Data is encrypted with a data key, and the data key is encrypted with the KMS key, which never leaves KMS. Controlling the KMS key controls everything it protects.
- AWS CloudHSM is a dedicated hardware security module for rules that say even AWS must have no access to the keys.
- S3 encryption. Every new object is encrypted by default. SSE-S3 uses keys S3 manages. SSE-KMS uses KMS keys, with an audit trail. SSE-C uses keys you supply and is off by default on new buckets since April 2026.
- AWS Certificate Manager (ACM) provides TLS certificates for encryption in transit. KMS keys encrypt stored data. ACM certificates encrypt data on the network. An answer that swaps them is wrong.
Protecting and governing data
- S3 Block Public Access stops buckets being made public by accident.
- S3 Versioning keeps every version, so an overwrite or delete can be undone.
- S3 Object Lock stores objects as write once, read many (WORM). In compliance mode not even the root user can delete them.
- AWS Backup manages backups across services, with copies to other Regions or accounts. AWS Backup Vault Lock stops anyone deleting them early.
- AWS CloudTrail records who made each API call. AWS Config checks whether resources meet a rule. AWS Artifact holds AWS's own compliance reports.
How Domain 1 questions are worded
| If the question says | Think |
|---|---|
| An application on EC2 needs access to another service | IAM role attached to the instance |
| Access between accounts without sharing credentials | Cross-account IAM role |
| Stop every account in an OU doing something | Service control policy |
| Fastest way to set up and govern many accounts | AWS Control Tower |
| Block one specific IP address | Network ACL |
| Private instances reach S3 without the internet | Gateway endpoint |
| Automatic rotation of a database password | AWS Secrets Manager |
| Control who can use the key and how it rotates | Customer managed KMS key |
| Who made this change | AWS CloudTrail |
| Do resources meet this rule | AWS Config |
| Suspicious activity in the account | Amazon GuardDuty |
| Software vulnerabilities on instances | Amazon Inspector |
Domain 2: Design Resilient Architectures (26%)
Domain 2 covers how applications scale, how you separate their parts, serverless and containers, how you keep running when something fails, and disaster recovery. Questions usually describe an application that falls over under load or when a server dies, and ask how to fix it.
Scaling and stateless servers
Vertical scaling makes one server bigger. It has a ceiling and usually needs downtime. Horizontal scaling adds more servers and has no real limit, so it is almost always the answer.
Horizontal scaling only works if servers are stateless. Move session data to a shared store such as Amazon ElastiCache or DynamoDB, so any server can handle any request.
Elastic Load Balancing
| Load balancer | Layer | Use it when |
|---|---|---|
| Application Load Balancer (ALB) | 7 (HTTP and HTTPS) | Web applications and microservices, routing by URL path or hostname |
| Network Load Balancer (NLB) | 4 (TCP and UDP) | Extreme performance, non-HTTP traffic or a static IP address per AZ |
| Gateway Load Balancer | 3 | Sending traffic through third-party appliances such as firewalls |
Amazon EC2 Auto Scaling
An Auto Scaling group launches instances from a launch template, keeps a minimum, desired and maximum number across AZs, and replaces any instance that fails its health check.
| Scaling type | Use it when |
|---|---|
| Target tracking | Keep a metric at a value, such as average CPU at 50%. The simplest and the one AWS recommends |
| Step scaling | Add or remove a set number depending on how far an alarm is breached |
| Scheduled scaling | Traffic spikes at a known time |
| Predictive scaling | Past traffic shows a pattern to scale ahead of |
A load balancer with an Auto Scaling group across multiple AZs is the standard answer for a scalable, highly available web tier. Note the names: Amazon EC2 Auto Scaling scales EC2 only. AWS Auto Scaling configures scaling for several resource types in one place.
Loose coupling
Loose coupling means parts of an application talk through something in the middle, so one failure does not spread.
| Service | What it does | Clue in the question |
|---|---|---|
| Amazon SQS | Message queue. A consumer pulls each message | Buffer work, decouple, absorb a spike |
| SQS FIFO queue | Strict order, exactly-once processing | Order matters |
| Amazon SNS | Publish and subscribe. Every subscriber gets a copy | Fan out one event to several systems |
| Amazon EventBridge | Event bus with routing rules and schedules | React to events from AWS services or SaaS applications |
| Amazon MQ | Managed ActiveMQ and RabbitMQ | Migrating an application that cannot be rewritten |
| AWS Step Functions | Orchestrates a multi-step workflow with retries | Steps, branching, waiting for approval |
| Amazon API Gateway | Front door for APIs | API keys, usage plans, throttling per client, request validation |
| AWS Transfer Family | Managed SFTP and FTPS into S3 or Amazon EFS | Partners must keep sending files over SFTP |
| Amazon AppFlow | Moves data between SaaS applications and AWS | Salesforce to S3 without code |
SQS details worth knowing: the visibility timeout hides a message while one worker processes it, a dead-letter queue catches messages that keep failing, and long polling cuts empty requests.
Retries, backoff and idempotency
When a service is being throttled, the answer is usually retries with exponential backoff and jitter, not a bigger backend. Operations that change data should be idempotent, so a retried payment request does not charge the customer twice.
Serverless and containers
- AWS Lambda runs code in response to events and bills per request and per millisecond. One invocation runs for a maximum of 15 minutes. Memory ranges from 128 MB to 10,240 MB, and CPU rises with memory.
- Amazon ECR stores container images. Amazon ECS is AWS's own container service. Amazon EKS runs Kubernetes.
- AWS Fargate runs containers on ECS or EKS with no EC2 instances to manage.
- AWS Elastic Beanstalk deploys a web application from uploaded code and manages the servers, load balancer and scaling.
For a job longer than 15 minutes where you still do not want to manage servers, choose containers on Fargate.
Highly available databases and storage
High availability means the system recovers quickly, with a brief interruption. Fault tolerance means no interruption at all, and costs more.
| RDS Multi-AZ | Read replicas | |
|---|---|---|
| Purpose | Availability | Read performance |
| Replication | Synchronous | Asynchronous |
| Failover | Automatic to the standby | Manual promotion |
| Other Region | No | Yes |
- Amazon Aurora keeps six copies of your data across three AZs and supports up to 15 read replicas. AWS says failover typically completes within 30 seconds. Aurora Global Database replicates to other Regions.
- Amazon RDS Proxy pools connections, which matters when many Lambda functions open connections at once.
- Amazon S3 is designed for 99.999999999% durability. S3 Cross-Region Replication copies objects to another Region.
- DynamoDB global tables replicate a table across Regions for local reads and writes.
To survive one AZ failing, Multi-AZ is enough. A second Region is the expensive answer for losing a whole Region.
Amazon Route 53 routing policies
| Policy | Use it when |
|---|---|
| Simple | One resource |
| Weighted | Split traffic by percentage, such as 10% to a new version |
| Latency | Send users to the Region that responds fastest |
| Failover | Switch to a backup site when the primary fails its health check |
| Geolocation | Route by the user's country or continent, for legal or language reasons |
| Geoproximity | Route by distance and shift traffic with a bias |
| Multivalue answer | Return several healthy addresses |
Disaster recovery
The recovery point objective (RPO) is how much data you can lose, measured in time. The recovery time objective (RTO) is how long you can be down.
| Strategy | What runs in the second Region | Recovery | Cost |
|---|---|---|---|
| Backup and restore | Backups only | Hours | Lowest |
| Pilot light | The core, usually the database, replicating. Servers off | Tens of minutes | Low |
| Warm standby | A smaller working copy of everything | Minutes | Medium |
| Multi-site active-active | The full system, serving traffic | Near zero | Highest |
Raise the service quotas in your standby Region in advance, or recovery fails when it tries to scale up.
Automation and visibility
- AWS CloudFormation defines infrastructure as code, so you can rebuild an identical environment in another Region.
- Immutable infrastructure means replacing servers with new ones built from an updated image, not patching them in place.
- Amazon CloudWatch collects metrics and logs and triggers alarms. AWS X-Ray traces one request through every part of an application.
- Legacy applications that cannot be changed: put them behind a load balancer in an Auto Scaling group across AZs, move the database to RDS Multi-AZ and add RDS Proxy.
The guide also lists managed AI services to use in place of building your own: Amazon Comprehend (text analysis), Amazon Polly (text to speech), Amazon Transcribe (speech to text), Amazon Translate, Amazon Textract (documents), Amazon Rekognition (images and video) and Amazon Lex (chatbots).
How Domain 2 questions are worded
| If the question says | Think |
|---|---|
| Requests are lost during a traffic spike | SQS queue between the tiers |
| One event must reach several systems | SNS fan-out to SQS queues |
| Messages must be processed in order | SQS FIFO queue |
| Traffic spikes at a known time | Scheduled scaling |
| The database must survive an AZ failure | RDS Multi-AZ |
| The database is slow under heavy reads | Read replicas |
| Too many database connections from Lambda | Amazon RDS Proxy |
| A job runs longer than 15 minutes, no servers to manage | Containers on AWS Fargate |
| The company already uses Kubernetes | Amazon EKS |
| RTO of hours on a tight budget | Backup and restore |
| The database must always be running, at low cost | Pilot light |
| Near-zero downtime across Regions | Multi-site active-active |
| Users must be served from a specific Region by law | Route 53 geolocation routing |
Domain 3: Design High-Performing Architectures (24%)
Domain 3 has the most services. It covers choosing the right storage, compute, database and network for a workload, and how you move, stream and analyse large amounts of data.
Block, file and object storage
| Type | What it is | AWS service |
|---|---|---|
| Block | A raw disk attached to one server | Amazon EBS |
| File | A shared file system many servers mount | Amazon EFS, Amazon FSx |
| Object | Objects stored by key and accessed over an API | Amazon S3 |
Amazon EBS volume types
An EBS volume lives in one AZ. A snapshot backs it up to S3 so you can restore it in another AZ or Region.
| Volume type | Use it for |
|---|---|
| General Purpose SSD (gp3) | Most workloads. 3,000 IOPS baseline, with IOPS and throughput added separately from size |
| Provisioned IOPS SSD (io2) | Demanding databases that need high, consistent IOPS |
| Throughput Optimized HDD (st1) | Large sequential workloads such as big data and logs, at low cost |
| Cold HDD (sc1) | Rarely accessed data, at the lowest cost |
Instance store is the fastest disk, but the data is lost when the instance stops. Use it only for temporary data such as caches.
Amazon EFS and Amazon FSx
| Requirement | Answer |
|---|---|
| Many Linux servers share the same files | Amazon EFS |
| Windows file share with SMB and Active Directory | Amazon FSx for Windows File Server |
| High performance computing or machine learning, linked to S3 | Amazon FSx for Lustre |
| The company already runs NetApp | Amazon FSx for NetApp ONTAP |
| Moving existing ZFS file systems | Amazon FSx for OpenZFS |
Amazon S3 performance and hybrid storage
- Multipart upload splits a large file and uploads the parts in parallel. AWS recommends it for files over 100 MB.
- S3 Transfer Acceleration speeds up uploads from distant users through the nearest edge location.
- Prefixes. S3 supports thousands of requests per second per prefix, so more prefixes means more total throughput.
- AWS Storage Gateway lets on-premises applications use cloud storage. S3 File Gateway presents a bucket as a file share. Volume Gateway backs up on-premises disks. Tape Gateway replaces physical backup tapes.
Choosing compute
| Instance family | Examples | Use it for |
|---|---|---|
| General purpose | M, T | Web servers and most applications |
| Compute optimised | C | Batch processing, video encoding, game servers |
| Memory optimised | R, X | In-memory databases, real-time analytics |
| Storage optimised | I, D | Very fast local disk access |
| Accelerated computing | P, G | Machine learning and graphics |
- Placement groups. Cluster packs instances together for the lowest latency. Spread puts each on separate hardware. Partition separates groups across racks for systems such as Hadoop and Kafka.
- User data is a script that runs at first launch. Instance metadata lets an application look up details about its own instance, including its role credentials.
- AWS Batch runs large numbers of batch jobs. Amazon EMR runs Apache Spark and Hadoop.
- Lambda gets more CPU when you give it more memory.
Choosing a database
| If the question says | Think |
|---|---|
| Relational, managed MySQL, PostgreSQL, MariaDB, Oracle, SQL Server or Db2 | Amazon RDS |
| Relational with higher availability and performance | Amazon Aurora |
| Unpredictable or occasional relational workload | Amazon Aurora Serverless |
| Key-value at massive scale, single-digit millisecond responses | Amazon DynamoDB |
| Microsecond reads from DynamoDB | DynamoDB Accelerator (DAX) |
| Cache in front of a database, or session store | Amazon ElastiCache (Valkey, Redis OSS or Memcached) |
| MongoDB compatibility | Amazon DocumentDB |
| Highly connected data, such as social graphs or fraud | Amazon Neptune |
| Apache Cassandra | Amazon Keyspaces |
| Analytics across billions of rows | Amazon Redshift |
Caching and database migration
- Lazy loading caches data only when it is requested and missing. The cache holds what people use, but data can go stale.
- Write-through updates the cache on every write. Data is fresh, but you cache things nobody reads.
- Time to live (TTL) expires cached data after a set time.
- AWS DMS moves a database to AWS while the source keeps running. A heterogeneous migration, such as Oracle to Aurora PostgreSQL, needs the schema converted first with the AWS Schema Conversion Tool or DMS Schema Conversion.
Edge networking
| Amazon CloudFront | AWS Global Accelerator | |
|---|---|---|
| What it does | Caches content at edge locations | Routes users over the AWS network to the nearest healthy endpoint |
| Caches | Yes | No |
| Traffic | HTTP and HTTPS | TCP and UDP |
| Static IP addresses | No | Two |
| Use it for | Websites, video, static files | Gaming, VoIP, fast failover between Regions |
CloudFront origin access control restricts an S3 bucket so users can reach files only through CloudFront. Geographic restrictions are the simplest way to block users in certain countries.
Connecting networks
| Need | Answer |
|---|---|
| Connect two or three VPCs | VPC peering (not transitive) |
| Connect dozens of VPCs, or VPCs plus on-premises | AWS Transit Gateway |
| Share one application with other VPCs | AWS PrivateLink |
Give every VPC and your on-premises network a non-overlapping CIDR range. VPCs with overlapping ranges cannot be peered.
Data ingestion, streaming and analytics
| If the question says | Think |
|---|---|
| Real-time stream processed by custom applications | Amazon Kinesis Data Streams |
| Load streaming data into S3 or Redshift with no code | Amazon Data Firehose |
| The company already uses Apache Kafka | Amazon MSK |
| Video from cameras | Amazon Kinesis Video Streams |
| Copy on-premises file data to AWS online | AWS DataSync |
| Run SQL on data in S3 with no servers | Amazon Athena |
| Convert formats or run ETL | AWS Glue |
| Control access across a data lake | AWS Lake Formation |
| Search and log analytics | Amazon OpenSearch Service |
| Dashboards for business users | Amazon Quick (now AWS Quick) |
Converting CSV files to Parquet with AWS Glue makes Athena queries faster and cheaper, because Athena charges by the data scanned.
Domain 4: Design Cost-Optimized Architectures (20%)
Domain 4 covers the tools that show where money goes, the ways to pay less for compute, the cheapest storage for how often data is used, database costs and network charges. Almost every question says "most cost-effective". The right answer is the cheapest option that still meets every requirement.
Cost management tools
| Question | Tool |
|---|---|
| Alert me before spending passes a limit | AWS Budgets |
| Show and forecast past spending | AWS Cost Explorer |
| The most detailed, line-by-line cost data | AWS Cost and Usage Report |
| What each team or project spends | Cost allocation tags |
| Which instances are the wrong size | AWS Compute Optimizer |
Consolidated billing in AWS Organizations combines usage across accounts, so volume discounts arrive sooner and Reserved Instance and Savings Plans discounts are shared.
EC2 purchasing options
| Option | Discount AWS quotes | Use it when |
|---|---|---|
| On-Demand | None | Short-term or unpredictable work that cannot be interrupted |
| Reserved Instances | Up to 72% | Steady usage of a known instance type for one or three years |
| Savings Plans | Up to 72% | Steady spend, with flexibility. A Compute Savings Plan also covers Fargate and Lambda |
| Spot Instances | Up to 90% | Work that can be interrupted with a two-minute warning, such as batch jobs |
| Dedicated Hosts | n/a | Software licences tied to physical servers, or compliance rules |
The pattern the exam likes is a mix: Reserved Instances or Savings Plans for the steady baseline, On-Demand for peaks and Spot for fault-tolerant extra work. The trap runs the other way too. If traffic drops to almost nothing overnight, a commitment pays for servers you do not need. The answer there is Auto Scaling or a serverless service.
Optimising compute
- Right-size instances to what the workload uses. AWS Compute Optimizer recommends sizes.
- AWS Graviton processors usually give better price performance than equivalent x86 instances.
- Switch off development and test environments outside working hours.
- Lambda costs nothing while idle, so it suits workloads that sit idle much of the time.
- Match availability to the workload. Production may need several AZs. A test environment can usually run in one.
- AWS Outposts puts AWS hardware in your own data centre for low latency or data residency.
S3 storage classes
| Storage class | Use it for | Retrieval |
|---|---|---|
| S3 Standard | Frequently accessed data | Immediate, no fee |
| S3 Intelligent-Tiering | Unknown or changing access patterns | Immediate |
| S3 Standard-IA | Accessed less than once a month, needed immediately | Immediate, fee per retrieval |
| S3 One Zone-IA | Infrequent data you could recreate | Immediate, one AZ only |
| S3 Glacier Instant Retrieval | Archives needed in milliseconds | Immediate |
| S3 Glacier Flexible Retrieval | Archives where waiting is fine | Minutes to about 12 hours |
| S3 Glacier Deep Archive | Long-term compliance archives | Up to 12 hours, or up to 48 hours for bulk |
- Minimum storage duration. Deleting early still costs the full period: 30 days in Standard-IA and One Zone-IA, 90 days in the two Glacier retrieval classes and 180 days in Deep Archive.
- S3 Lifecycle rules move objects to cheaper classes and delete them automatically.
- Requester Pays makes the person downloading the data pay for the transfer.
Other storage and databases
- Move EBS volumes from gp2 to gp3, and delete unattached volumes and old snapshots.
- Use EFS lifecycle management to move unused files to the Infrequent Access and Archive classes.
- Use AWS Backup retention rules so old backups are deleted.
- DynamoDB on-demand suits unpredictable traffic. Provisioned capacity with auto scaling is cheaper for steady traffic.
- Aurora Serverless suits databases used occasionally. Reserved Instances cut the cost of RDS that runs all the time.
- A cache can be cheaper than a larger database or more read replicas when the same reads repeat.
- Moving from a commercial engine such as Oracle to Aurora PostgreSQL removes licence costs.
Network and data transfer costs
| Traffic | Charged |
|---|---|
| Into AWS from the internet | No |
| Out of AWS to the internet | Yes |
| Between AZs | Yes |
| Between Regions | Yes, at a higher rate |
| Within one AZ over private IP addresses | No |
| From an AWS origin to CloudFront | No |
- NAT gateways charge per hour and per gigabyte processed. One shared NAT gateway is cheaper but is a single point of failure, and other AZs pay cross-AZ charges to reach it. One per AZ costs more and is highly available.
- Gateway endpoints for S3 and DynamoDB are free and remove the NAT processing charge on that traffic.
- VPC peering has no hourly charge. Transit Gateway charges per attachment and per gigabyte, so peering is cheaper for a few VPCs.
- Site-to-Site VPN is cheap to run. Direct Connect costs more to set up but lowers the per-gigabyte cost for large, steady volumes.
How Domain 4 questions are worded
| If the question says | Think |
|---|---|
| Steady usage for a year or more | Reserved Instances or Savings Plans |
| Flexibility to change instance types or move to containers | Compute Savings Plan |
| Workload can be interrupted | Spot Instances |
| Licences tied to physical servers | Dedicated Hosts |
| Unknown access pattern for S3 data | S3 Intelligent-Tiering |
| Recreatable data at the lowest cost | S3 One Zone-IA |
| Archive kept for years, hours to retrieve is fine | S3 Glacier Deep Archive |
| Cheapest way for private instances to reach S3 | Gateway endpoint |
| Cut data transfer costs for a global website | Amazon CloudFront |
| Reduce NAT gateway costs | Keep traffic in the same AZ, or take the NAT gateway out of the path |
Four mistakes to avoid when you study for SAA-C03
- Using outdated material. AWS revised the exam guide in August and October 2026, removed Elastic Transcoder and uses new service names. Check that your course and practice questions match the current guide.
- Studying everything at equal depth. Security and resilience are 56% of the scored content. Spend time in proportion to the weights, then on your weakest areas.
- Studying passively. The exam gives you four answers that would all work. Telling SQS from SNS, Multi-AZ from a read replica, or CloudFront from Global Accelerator takes active recall: testing yourself until you can produce the answer from memory.
- Having no way to measure readiness. People book too early because they feel confident, or keep delaying because they never feel ready. Both come from not measuring what you know in each domain.
Exam day tactics
- Read the last line first. The scenario is mostly background. The final sentence says what is being asked, such as the most cost-effective option.
- Eliminate by requirement. Each scenario has two or three hard requirements. Cross out any answer that breaks one. This also works on services you have not studied.
- Watch the qualifier.
| Qualifier | It usually points to |
|---|---|
| Most cost-effective | Spot, Savings Plans, a cheaper storage class or a gateway endpoint |
| Least operational overhead | A managed or serverless service such as Lambda, Fargate, Aurora or DynamoDB |
| Highly available | Multiple Availability Zones |
| Disaster recovery | Multiple Regions |
| Least privilege | A role with a policy scoped to the one resource |
What to do after the course
AWS does not publish recommended study hours or pass rates, so the order below is my own approach.
- Read the official exam guide and mark each task statement strong, shaky or unknown.
- Take a diagnostic by domain, so you do not spend time on what you already know.
- Go back over the domains by weight: Domain 1 (30%), Domain 2 (26%), Domain 3 (24%), Domain 4 (20%).
- Test yourself after each topic and keep every wrong answer to review again.
- Sit a full-length timed mock: 65 questions in 130 minutes.
- Book when your scores are steady in all four domains on questions you have not seen before.
Prepare with StudyTech AI
StudyTech AI is a personalised AWS and AI certification prep platform. It finds your knowledge gaps, builds your practice around your skills and gaps, turns your mistakes into flashcards, scores your readiness for every exam domain, and tells you when you are ready to take your exam.
Pricing is $29 a month or $297 a year, and the diagnostic is free.
- Try 15 free Solutions Architect Associate practice questions
- Check whether you are ready for SAA-C03
- See how readiness is scored on the AWS exam readiness page
- Browse all AWS practice exams on StudyTech AI
Start a free diagnostic on StudyTech AI
StudyTech AI is not affiliated with or endorsed by Amazon Web Services.
Cite this page
StudyTech AI. "AWS Solutions Architect Associate Full Course 2026 (SAA-C03)." studytech.ai/blog/aws-solutions-architect-associate-full-course-2026. Updated 16 November 2026.
Related reading
- AWS Solutions Architect Associate study guide
- SAA-C03 domain breakdown
- AWS Solutions Architect Associate exam format
- Is the AWS Solutions Architect Associate worth it?
- AWS Solutions Architect Associate certification page
- AWS certification roadmap 2027
Verified 16 November 2026 against the official AWS Certified Solutions Architect - Associate (SAA-C03) exam guide (version 1.1), its four domain pages, the in-scope services list, the revisions page and the AWS certification page. AWS can update these pages, so confirm the current version before you book.